Salvato in:
Dettagli Bibliografici
Autori principali: Dong, Chengqi, Yue, Chuhuai, He, Hang, Mao, Rongge, Tang, Fenghe, Zhou, S Kevin, Xu, Zekun, Wang, Xiaohan, Chai, Jiajun, Yin, Guojun
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2512.08980
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917381510529024
author Dong, Chengqi
Yue, Chuhuai
He, Hang
Mao, Rongge
Tang, Fenghe
Zhou, S Kevin
Xu, Zekun
Wang, Xiaohan
Chai, Jiajun
Yin, Guojun
author_facet Dong, Chengqi
Yue, Chuhuai
He, Hang
Mao, Rongge
Tang, Fenghe
Zhou, S Kevin
Xu, Zekun
Wang, Xiaohan
Chai, Jiajun
Yin, Guojun
contents Recent VLM-based agents aim to replicate OpenAI O3's "thinking with images" via tool use, yet most open-source methods restrict inputs to a single image, limiting their applicability to real-world multi-image QA tasks. To address this gap, we propose IMAgent, an open-source visual agent trained with end-to-end reinforcement learning for fine-grained single/multi-image reasoning. During inference, VLMs tend to gradually neglect visual inputs; to mitigate this issue, we design two dedicated tools for visual reflection and verification, enabling the model to actively refocus attention on image content. Beyond that, we, for the first time, reveal how tool usage enhances agent performance from an attention perspective. Equipped with a carefully designed two-layer motion trajectory masking strategy and tool-use reward gain, IMAgent acquires an effective tool-use paradigm through pure reinforcement learning, eliminating the need for costly supervised fine-tuning data. To further unleash the inherent tool-usage potential of the base VLM and fill data gaps, we construct a challenging, visually enriched multi-image QA dataset via multi-agent system. Extensive experiments validate that IMAgent achieves SOTA performance across mainstream single and multi-image benchmarks, and our in-depth analysis offers actionable insights for the community. Code and data will be released soon.
format Preprint
id arxiv_https___arxiv_org_abs_2512_08980
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Training Multi-Image Vision Agents via End2End Reinforcement Learning
Dong, Chengqi
Yue, Chuhuai
He, Hang
Mao, Rongge
Tang, Fenghe
Zhou, S Kevin
Xu, Zekun
Wang, Xiaohan
Chai, Jiajun
Yin, Guojun
Computer Vision and Pattern Recognition
Artificial Intelligence
Recent VLM-based agents aim to replicate OpenAI O3's "thinking with images" via tool use, yet most open-source methods restrict inputs to a single image, limiting their applicability to real-world multi-image QA tasks. To address this gap, we propose IMAgent, an open-source visual agent trained with end-to-end reinforcement learning for fine-grained single/multi-image reasoning. During inference, VLMs tend to gradually neglect visual inputs; to mitigate this issue, we design two dedicated tools for visual reflection and verification, enabling the model to actively refocus attention on image content. Beyond that, we, for the first time, reveal how tool usage enhances agent performance from an attention perspective. Equipped with a carefully designed two-layer motion trajectory masking strategy and tool-use reward gain, IMAgent acquires an effective tool-use paradigm through pure reinforcement learning, eliminating the need for costly supervised fine-tuning data. To further unleash the inherent tool-usage potential of the base VLM and fill data gaps, we construct a challenging, visually enriched multi-image QA dataset via multi-agent system. Extensive experiments validate that IMAgent achieves SOTA performance across mainstream single and multi-image benchmarks, and our in-depth analysis offers actionable insights for the community. Code and data will be released soon.
title Training Multi-Image Vision Agents via End2End Reinforcement Learning
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2512.08980